Champagne Jin

Thanks for your work. It helps a lot : ) Could you please add the the derivation process of 'KL divergence between 2 Laplace Distribution'? Thanks you again.

I use `tw.draw_model` to visualize my model. Thank you so much.

Supplementary material is not available

4

Thanks for your job! As the title says, however, i cannot download the supplmentary material in your project page anymore. Could you update it or realease a new link?

Question about 300W-LP training

1

Thanks again for your work. Can you please provide the training details of the the MOS-S-300wlp model? How did you define the supervision, and what are your training settings?

Thanks again for your work. Since you use flip and multi-scale strategy in testing, I wonder how your model performs when these test tricks are not applied. I ask this...

BatchNorm1d 统计参数出现NaN的问题

1

你好，作者。 在你的代码中，有两处使用了BatchNorm1d，而我在训练过程中发现，随着迭代次数的增加，其统计参数在未知原因下会变成NaN，由于BN的统计参数不影响训练过程，因此训练完全没有异样，但在测试时就会出现问题。 根据我的检查，保存的模型参数中，有且只有BatchNorm1d的统计参数出现了NaN的问题。 请问你是否遇到这类情况？（我可以确保输入不是全0，batchsize不是1）